Architecture¶

Two backends. kwiloai_webapp serves Campus, Classroom and Compliance; kwiloai-b2c serves the consumer Learner product with its own database. Both share the chatbot, eval service, learner memory and the question bank. Clients never talk to a database, AI provider or payment gateway directly.

flowchart TD
  subgraph Clients
    SITE["Marketing site<br/>kwilo.ai · apps/site"]
    WEB["Product app<br/>app.kwilo.ai · apps/web"]
    CAMPUS["Kwilo Campus<br/>apps/mobile-b2b"]
    LEARN["Kwilo Learn<br/>kwiloai-b2c"]
  end
  SITE --> WEB
  WEB --> API
  CAMPUS --> API
  LEARN --> B2C
  B2C["Consumer API<br/>kwiloai-b2c · own DB"]
  B2C --> BOT
  B2C --> BANK
  B2C --> MEM
  API["Backend API<br/>api.kwilo.ai · apps/backend<br/>FastAPI on Azure Container Apps"]
  API --> DB[("Azure Postgres<br/>app DB + erp schema")]
  API --> MEM[("kwiloai_memory<br/>learner profiles")]
  API --> BANK[("knowledge_base_status<br/>question bank, textbooks")]
  API --> BOT["Chatbot microservice<br/>all LLM work"]
  API --> EVAL["Eval service<br/>solver, heavy jobs"]
  API --> EXT["Razorpay · MSG91 · Azure Blob · email"]

Apps and packages¶

Path	What it is	Hosted
`apps/site`	Public marketing site, prerendered HTML for SEO. Deliberately lean: no TanStack Query, Zustand or TipTap.	Cloudflare Pages, `kwilo.ai`
`apps/web`	Product app for every signed-in role. React 19, Vite, Tailwind 4, TanStack Query, Zustand.	Cloudflare Pages, `app.kwilo.ai`
`apps/backend`	FastAPI, async SQLAlchemy 2, Alembic. Includes the ERP under `src/erp`.	Azure Container Apps, `api.kwilo.ai`
`apps/mobile-b2b`	Kwilo Campus, In build. Expo, expo-router.	Google Play, via `mobile-b2b-play.yml`
`MySetu-AI/kwiloai-b2c`	Consumer backend, web and the Kwilo Learn mobile app. Production cutover In build.	Azure Container Apps, Cloudflare Pages
`apps/workers/ph-relay`	Cloudflare Worker relaying PostHog traffic	Cloudflare
`@kwilo/ui`, `@kwilo/ui-native`, `@kwilo/core`	Design system, consent, analytics and auth contract. Published from `MySetu-AI/kwilo-ui` as `@mysetu-ai/{web,mobile,core}` and installed under the `@kwilo/*` aliases.	GitHub Packages

Rule of placement: anything auth-protected goes in apps/web. Anything public and crawlable goes in apps/site.

Pages¶

Page	Covers
Auth and access	Login, tokens, roles, capabilities, B2C vs institutional identity
Data and AI	Databases, the chatbot and eval services, retrieval, the question bank
Infrastructure and deployment	Environments, Azure, Cloudflare, secrets, releases
Analytics agent	Ask Kwilo isolation walls, tools, config
Analytics and error tracking	PostHog events, replay, errors
Attribution and UTM	The site → app → backend attribution contract
Kwilo Learn app	Builds, env vars, distribution for the consumer Android app
App conventions	Each app's `AGENTS.md` in the code repo

Principles¶

Backend is the policy layer. Auth, roles, quotas, entitlements, persistence and AI labelling live in the backend.
Chatbot is the LLM layer. Prompts, generation, retrieval and token usage live in the chatbot service. The backend holds no generation prompts.
No heavy compute in the request path. Long jobs like timetable solves go to the eval service through a queue.
Tenant isolation in the database. ERP tables use forced row-level security keyed on the active institution.